Change GeometryDtype.na_value to None

[jorisvandenbossche]

The GeometryDtype.na_value in practice is None (this is being returned when accessing a missing value), but the attribute itself it set to NaN.

In the past I remember that it gave some test failures, but let's see what it does now.

[jorisvandenbossche]

And indeed, it gives a whole bunch of errors, it seems up to pandas 2.0. So we might be able to change this in the future when supporting pandas >= 2.0

[m-richards]

And indeed, it gives a whole bunch of errors, it seems up to pandas 2.0. So we might be able to change this in the future when supporting pandas >= 2.0

I imagine defining na_value conditionally based on pandas version would be bad? (I could imagine this might be a problem for pickling but not sure)

[jorisvandenbossche]

It might not be too bad to make na_value dependent on the pandas version, but I would personally like to avoid that I think

[m-richards]

With pandas-dev/pandas#54930 this is no longer needed, and I suppose we can close this. (Though longer term it's perhaps still worth looking at setting the NA value to None for consistency once we only support pandas >=2)

[m-richards]

Merged this with main again now that we support pandas >=2 only. Seems we may as well just do this for consistency, since the behaviour is that the na_value is None in practice (and maybe that let's the pandas side simplify tests if they want to)